Greetings from a first time (diary) caller. I've been skeptical about the idea that this election was fraudulently stolen. Not because Bush, Rove, and Co wouldn't stoop that low, but rather because the amount of fraud needed to produce these election results seems too large and widespread. Crooked yes, but stupid? Getting caught stealing the election by manipulating hundreds of thousands of votes would be political suicide for the GOP with lasting effects. Better to lose than to face such recriminations.
But systematic "fraud" did occur. The FL felons list, the Sproul Assoc. shenanigans, the Ohio vote machine deployment, are all examples of election "fraud" in my opinion, perhaps with the tacit knowledge by the national party. This sets the scale for fraud: a few thousand votes here and there, enough to swing a razor thin election or make a small lead insurmountable (Ohio?), but still be under the radar. If exposed, be deniable: attributed to a few bad apples - easy in our media climate.
With that `scale' in mind, the results of the Miami Herald Recount might just suggest that something is amiss.
As you're probable aware, the Miami Herald performed a hand recount in 3 northern Florida counties, a complete recount in 2 counties and a partial recount in a third. Their goal was to determine whether the large mismatch between the Democratic registration numbers and the Bush vote in northern Florida was indicative of some massive fraud. Their hand recount `sufficiently' matched the initial count for them to conclude that no fraud occurred.
This recount has been discussed here at DKos before. This
recent diary wrote that the results from the partially recounted county showed a significant discrepancy from the initial tally. While very interesting, it is not conclusive since, as noted in that diary, we don't know how well the partial recount sampled the county. The difference just might be explainable. What caught my attention in the MH report, though, were the results from the 2 counties where the full recount was performed.
One person, not swayed by the above mentioned diary, commented there,
The two counties they hand counted in their entirety showed no real difference.
Well, I'm not convinced about this. Let me first reprise the MH results:
County |
Bush vote |
Bush recount |
Bush change |
Kerry vote |
Kerry recount |
Kerry change |
% change towards Bush |
Union |
3396 |
3393 |
-3 |
1251 |
1272 |
+21 |
-0.52% |
Lafayette |
2460 |
2452 |
-8 |
845 |
848 |
+3 |
-0.33% |
Suwanee |
11153 |
6140 |
-- |
4522 |
2984 |
-- |
-7.4% +/- ? |
The % change is defined as net-change/total-votes expressed as a percentage. The last county is an extrapolation and does not represent the actual change. But from the 2 completely recounted counties, I make these observations:
- Bush won both counties convincingly
- the recount changed the vote in Kerry's favor in both counties
- Kerry gained votes in both counties
- Bush lost votes in both counties
I can imagine many are unimpressed at this small effect. Surely there are those of you who know binomial statistics screaming, "wait! wait! wait! with an ~8000 vote sample in the 2 counties we expect an `uncertainty' of about 1%, much bigger than the ~0.5% change you observe - nothing interesting here". Hmmm, how to put this gently, NO NO NO NO NO! Such statistics work under specific conditions. Let's say you have a large parent sample of B&K votes and from this sample you select 8000 votes. Then you select another 8000 vote sample, and another and another.... Now when you compare the results between your samples, you'll find that they differ, on average, by about 1% as described by binomial statistics - this is what we mean by an uncertainty when taking only one 8000 vote sample. But this is NOT what is done in a recount. Here they are recounting every ballot in the original sample. Throw the stat-models out the window; they don't apply. The vote count can change for a variety of reason, the counting process, the technology used, rate of ballot spoilage, even fraud, but there is no way to tell whether such a change is big or small without re-running the experiment. You have to count many more counties or compare with some historical results in order to have any idea whether a 0.5% change is reasonable or not. Frankly, such a vote change seems a bit large to me.
One of the reasons this caught my attention was a comment I heard here in Washington State recently from an election official about recounts. She said in places where the winner is clear, recounts tend increase the winner's lead. Simply put, random mistakes will tend to affect the votes in proportion to the original count. My first two observations from the MH recount find this "rule-of-thumb" violated.
Ok, as I said we cannot know how "big" this change really is without some other comparisons, so I decided to take a look at the results here in Washington. First I note that our election law requires a recount at taxpayer expense in any state-wide race where the vote difference is less than 2000 votes. Since we're paying for it, I assume this statute is based on some historical information. It implies that the original tally is reliable to ~2000 votes out of about 2 million or 0.1%. This is 3-5 times smaller (more reliable) than found in the MH recount. Well, we had a recount here in the Governor's race and the results differed by about 200 votes, a change of less than 0.01% or 30-50 times smaller than the MH recount. Hmmm.
We can also look at the county-by-county details of the WA Governor's recount. Now note, this was a machine recount not a hand recount so it is not completely comparable to the hand recount in Florida, but it's what I got.
First let's consider my 4th observation: Bush lost votes in both FL counties. Is that odd? In the WA recount we find 8 out of 39 counties where one or both candidates lost votes. In four of those, one candidate lost just 1 vote. In two of those counties, both candidates lost a handful of votes which to me suggests a singular event (double counts in the original tally, damaged or lost ballots, ...) caused such problems. Only in 2 counties did one candidate lose several votes like we see in the FL recount. In any event, we find that 15%-20% of the counties had vote losses in the WA recount. In FL we were 2-for-2. In those last two WA counties, the vote loss (per vote-total of the candidate that lost votes) was at a rate of 0.03% and 0.05% compared to 0.09% and 0.3%.in the two FL counties. The FL recount vote-loss rate was a factor of 2-to-10 larger than the WA recount..
We can also check out our `rule-of-thumb'. I found 31 WA counties where there was a winner and a vote change in the recount. Twenty-two of these 31 changed in favor of the county's winner, 9 changed towards the loser. Thus we find the `rule' holds at about a 2:1 rate. In FL, we again were 2-for-2 the other way.
Finally, let's consider how big the changes were in the WA recount. Not big at all. In 37 of the 39 counties, the change was less than 0.1% and most counties changed by less than 0.03%. There was 1 county with a change comparable to the Fl recount; Adams county changed by 0.3%, a small county where Rossi (the Rep) won big (70%:30%) and the recount change went in Rossi's favor. The Adams result also showed an increased vote count for both candidates at a rate more than twice that of any other county, suggesting they found some missing ballots. These additional votes were in the same proportion (70%:30%) as the initial count which completely explains this apparent outlier. This is also what you might expect if you accepted votes that the machine could not decipher, just the opposite of what we see in the MH recount.
I think the best way to show the scale of these changes is graphical. Below is a histogram of the %change times the direction of the change. The direction is either 1 or -1 depending on whether the change increased the winner's vote in that county or that of the loser. The 2 FL counties are shown in blue. The Adams Co. result is the one at the far right.
What we see is that the 2 FL recounts are BOTH clearly outliers with respect to the WA recount distribution. What does this mean? Well, I guess it could be a fluke. Or perhaps the FL vote isn't very reliable so that if we counted more counties we'd find just as many where Bush gained votes at a 0.5% rate. Or it could suggest that somehow they were flipping votes (Kerry to Bush) at a rate of 1 vote per ~400. If done over, say, half the state, this would translate to maybe a 25,000 vote differential. In the context of the FL 2000 election, this would be decisive. Did they have the access to commit such a fraud? - a good question to have answered.
The major limitation with this analysis is that it only covers 2 counties - not a sample size I want to go screaming fraud over. One reason I am writing this diary is for more people to pay attention to these details. If a recount comes in that changes the results by only a handful of votes, don't assume that it is meaningless. It might well be significant. Finally, in the context of this election, finding fraud that shifts 0.5% of the vote will not change the results. It might, however, change the context and oversight of the political debate during the next four years.
Some random comments:
- I saw in a comment thread that the New Hampshire recount is done. Does anyone know if the data is available?
- It may be that the reporters doing the FL recount were bored, tired, and sloppy. They were after all looking for a significant swing.
- I sent these observations a few days ago (via email) to one of the reporters of the MH story. I suggested recounting a couple more counties and/or comparing their results with the 2000 hand recount. I have not received a response.